RE: LeoThread 2025-05-06 15:25
You are viewing a single comment's thread:
Observing a distinct pattern in the approach to AI-assisted coding—focusing on work that requires professional care rather than casual experimentation—the process is structured as follows:
Step 1: Compile all relevant context.
0
0
0.000
This might be time-consuming for large projects, but for smaller ones, simply aggregating everything (for instance, using a command like "files-to-prompt . -e ts -e tsx -e css -e md --cxml --ignore node_modules -o prompt.xml") is
sufficient. Step 2: Clearly outline the specific incremental change to be implemented. Instead of asking for code right away, request several high-level strategies, weighing their pros and cons.
Multiple methods often exist, and the AI's judgment might not always align with the best approach; therefore, it's useful to add concrete details when possible. Step 3: Choose one strategy and request initial draft code based on that method.
Step 4: Enter a review and learning phase by manually consulting relevant API documentation for any unfamiliar functions. Request clarifications or alternative explanations if needed, and then adjust or consider a different approach.
Step 5: Test the resulting code thoroughly. Step 6: Commit the changes using version control. Finally, seek suggestions for the next implementation and repeat the cycle.
This approach mirrors a refined inner loop in AI-assisted development, emphasizing a careful balance between leveraging the AI's vast knowledge and maintaining strict control over code quality.
The focus remains on being methodical, cautious, and engaging in continuous inline learning rather than simply deferring to the AI’s expertise.
Although many stages currently involve a manual or cumbersome process, there is significant potential for improved UI/UX and better tool support as this field continues to evolve.